Abstract
This thesis is based on using stochastically resonant spectrum sensing (SRSS) to
enhance the sensing and detection of weak radio frequency (RF) signals by a cognitive
radio (CR) in dynamic spectrum access (DSA) networks (DySPAN), e.g., television
whitespace (TVWS), or even across the spectrum so to speak. Sharing spectrum
dynamically implies a consequent increase in interference that must be tolerated to
some extent given the nature of opportunistic DSA, and which in turn increases the
background noise floor, below which threshold signals remain essentially undetectable,
or such interference can be managed, or even be harnessed or used to advantage.
The thesis hypothesises to exploit stochastic resonance to enhance the sensing and
detection of weak RF signals by a DySPAN CR, which explicitly accounts for and
includes, propagation impacts, the prevailing background radio noise and a multi-user
interference (MUI) component, that is fed by different stochastically resonant noise
(SRN) distributions (SRNDs) representing, mimicking, modelling or characterising
different types of radio environments (REs). This implies not only explicitly accounting
for the impact of the RE through propagation impacts, background noise and prevailing
MUI in the SRSS detection enhancement process, but also using the RE as characterised
by alternative SRNDs to generate the input SRN to inject.
Typically, the injected SRN is additive white Gaussian noise (AWGN), which is
normally distributed, and this thesis uses ten different SRNDs to characterise alternative
REs as follows: logarithmic for analytic purposes; uniform for such band-limited
pseudo white-noise-like environments; normal/Gaussian which is typically or normally
used; Rician for line-of-sight propagation; Rayleigh for non-line-of-sight propagation
with multiple reflections; Nakagami to generalise both Rician and Rayleigh
environments; lognormal for multi-path fading environments, and beta, gamma and
exponential for theoretical and mathematical analysis.
In effect, the thesis probes the prospects of not only including the RE in the SRSS
process itself but also interconnecting the RE as characterised by SRNDs and SRSS in
a feedback loop to assist in the sensing and detection of weak RF signals and managing
increasing interference as would occur and proliferate in DSA/DySPAN/TVWS. This
bodes well for employing SRSS in a DySPAN CR since it implies not only using, e.g.,
mitigating, harnessing, harvesting, exploiting, etc., any interference, i.e., interference
management (IM), in enhancing the sensing and detection of weak RF signals in
DySPAN, it also models a complete RE comprising of the wanted signal, the prevailing
RE noise of the propagation channel and a combined MUI component representing all
unwanted interference, and, thereby enabling more accurate RE monitoring and
mapping (REMM) at the same time. The RE is thus embedded, or rather embroiled, in
the SRSS sensing and detection enhancement, and results in a feedback loop.
Now, to more fully investigate the SRSS process in detail and to better explore the
mixing in of the RE when used as the injected SRN generated by some SRND, the
thesis goes beyond the typical case usually employed, an additive SRN (ASRN-ONLY)
model, to also include multiplicative SRN (MSRN) and as a result considers various
combinations of ASRN-MSRN, resulting in eighteen SRSS models, including,
SINGLE-ASRN-ONLY, DUAL-ASRN-ONLY, combined ASRN-MSRN, MSRNONLY,
a reference model without any injected SRN, and a novel set of NRSS-SRSV
models that breaks traditional modelling in trying to reduce variables by replacing the
normalised received signal strength (NRSS) for the SR state variable (SRSV) response.
vi
The research methodology is based on modelling and a SRSS simulation (SRSSS) in
MATLAB. For each SRSS model, the applicable SR stochastic differential equation
(SDE) is obtained, solved using various numerical integration techniques to check and
verify nonlinear effects, uses a fortified version of MATLAB’s ode45 to ensure the
final nonlinear numerical integration is stable, and obtains a result for the SR SDE state
variable response, which are then plotted against the injected SRN amplitude (SRNA)
and presented, discussed and analysed, allowing the responses of different SRSS
models to be compared to and contrasted against each other.
Detailed analysis of the ASRN-ONLY model is conducted using established theory and
these results are extended to explicitly account for MUI versus NO MUI, as well as
subjecting the ASRN-ONLY SRSS model to the ten SRNDs described above, for which
a SR output signal-to-noise ratio (SNR) is obtained, and from which a corresponding
SR output-to-input SNR improvement is computed. A SR input scaling parameter
scales the SR input signal, is set at a typical value and also is varied over a range of four
fixed values, SR SDE parameters are fixed, and the SR state variable response results
for all SRSS models are presented for either, an illustrative and theoretical logarithmic
range of monotonically increasing SRNAs for the four values of the SR scaling
parameter, or for all ten SRNDs for a typical value of the SR scaling parameter.
Post-SR processing of the SR SDE state variable response results for all eighteen SRSS
models includes determining a representative average (AVE), the energy (END) in the
response, a maximum (MAX) response and the corresponding SRNA (MAX_ID) for
all ten SRNDs for a typical value of the SR scaling parameter, and a maximum response
(MAX) and the corresponding SRNA (IDX) for the illustrative logamp SRND for four
values of the SR scaling parameter.
SR steady-state signal and noise responses for the ASRN-ONLY model for a typical
value of the scaling parameter for all ten SRNDs are then estimated using established
theory, providing an applicable threshold (gam-sr-ed), against which an energy
detection statistic (T-X) can be computed, in two ways, using centralised or
standardised (subtract the mean and divide by the standard deviation) values of the
representative average energy (AVE-END), as well as the actual energy (END), adding
the impact of MUI so that it can be compared to NO MUI.
A final energy detection test statistic result (T-X-AVE-END, T-X-END) using the two
energies (AVE-END, END) is then performed for all SRSS models using the ASRNONLY
SRSS model as a reference (gam-sr-ed) model.
The results are presented, discussed and analysed, from which the research findings are
extracted and drawn. The research findings are then tabled, beginning thematically, to
generally, and finally honing in specifically, to establish the thesis. Final
recommendations are proffered, and the thesis conclusions are codified, posed and
tendered.